[c-nsp] RBE DSL

[Andre Beck]

Hi Scott,

just in case there really were no answers:

On Tue, Jun 07, 2005 at 12:52:07PM -0700, Voll, Scott wrote:
> I'm setting up RBE for the first time for an internal DSL network.
> 
> I'm using our CAT 6509 and an 837 at the subscriber site.  What does the
> ip addressing look like with RBE?
> 
> A normal network?  1.1.1.x/24------- Cat 6509 ------172.16.1.1/30 ------
> ATM ------ DSL ---- 172.216.1.2 ------- 837 ---- 1.1.2.x/24

If you replace the 172.216.1.2 with 172.16.1.2 then yes, this is
a possible setup. You just deal with an RBE interface as with any
routed interface towards a broadcast domain.
 
> Or do I share the same network at both sides of the ATM/DSL?

No. RBE is a routed interface. The idea is to hide the fact that the
endpoints of the ATM PVC are actually used to build a AAL5-SNAP to Ether-
net Remote Bridge from the configuration by automatically topping the
bridge with an IP L3 interface. Primary reason for that is that it would
by no means scale to do the same thing using IRB, BVIs and real explicit
bridge-groups. Internally, it is of course never bridging but just
encapsulating the packets in the way expected by the other side, and
that is the reason why it is more efficient. The only real bridge in-
volved is the one used at the CPE side for DSL deployment on Ethernet.

Now, we are at the real interesting question: Why in the world would
you use RBE *at* *all* when you are able to terminate it directly using
an ADSL WAN port equipped router? The whole thing is necessary only when
there is a bridge involved at the far side, usually as the carrier-
installed CPE which supplies 10BaseT/100BaseTX to the customer's own
hardware. And normally this is done by the carrier to supply PPPoE
only. On the "naked" DSL, you have direct access to the ATM PVC, making
AAL5-SNAP and RBE a useless waste of bandwidth and ressources. Just use
an unnumbered subinterface linked to an AAL5-MUX PVC for IP only and you
double your throughput (saleswise, actually only for TCP ACKs, but he).

Well of course, the answer depends on what router exactly you are going
to use. Most cheap-sold crap on that market will probably not even know
that anything else but SNAP exists. So deploying RBE is mostly a question
of uniformity for large scale providers. If you are doing that for your
own purposes or with flexible customers and have your own choice of CPE
hardware, well, see above.

> Sorry,  I'm a little confused.

Layer-mixups are always confusing. Just try to build the same thing
using IRB[1], if you get *this* you will welcome how easy, clean and
transparent RBE is ;)

[1] I know I'm bad. Pointing people who suffer a layer mixup into the
    direction of Cisco IRB documentation is probably violating the
    Geneva convention.
-- 
                  The _S_anta _C_laus _O_peration
  or "how to turn a complete illusion into a neverending money source"

-> Andre Beck    +++ ABP-RIPE +++    IBH Prof. Dr. Horn GmbH, Dresden <-